Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017
Enhancement of Twins Fetal ECG Signal Extraction Based on Hybrid Blind Extraction Techniques Ahmed Kareem Abdullah Hadi Athab Hamed AL-Musaib Technical College, Al-Furat Al-Awsat Technical University
[email protected] [email protected] Nasser Ali Hasson AL-Zubaydi AL-Musaib Technical Institute, Al-Furat Al-Awsat Technical University
[email protected]
Abstract ECG machines are noninvasive system used to measure the heartbeat signal. It’s very important to monitor the fetus ECG signals during pregnancy to check the heat activity and to detect any problem early before born, therefore the monitoring of ECG signals have clinical significance and importance. For multifetal pregnancy case the classical filtering algorithms are not sufficient to separate the ECG signals between mother and fetal. In this paper the mixture consists of mixing from three ECG signals, the first signal is the mother ECG (M-ECG) signal, second signal the Fetal-1 ECG (F1-ECG), and third signal is the Fetal-2 ECG (F2-ECG), these signals are extracted based on modified blind source extraction (BSE) techniques. The proposed work based on hybridization between two BSE techniques to ensure that the extracted signals separated well. The results demonstrate that the proposed work very efficiently to extract the useful ECG signals. Key words: ECG, Blind source extraction, blind source separation, Twins Fetal ECG
1. Introduction The heartbeat of the fetus are very fast from mother’s heartbeat about 120-160 beat/minute, but the amplitude of the ECG-fetus is very weak compared with motherECG signal (Álvarez and Alberto, 2015). The technique for getting the fetus ECG signals are noninvasive techniques (Gao et.al., 2003 ). The electrical signals generated from fetus heart are very small and weak. The position of the electrodes are very important to detect useful information from the mixture (Saranya and Priyadharsini 2010), therefore the electrodes or sensors are sited on the abdomen as near as possible to the heart of the fetus. The ECG signals are collected by putting the electrodes about the place of the fetal and wanting the electrodes have the high signal to noise ratio(Saranya and Priyadharsini, 2010). The position of electrodes represent one of the problem can be solved by expert but the noise and interface with ECG signal still effect on the diagnosis. Many methods like doppler ultrasound used to detect fetus ECG signal during 10th-12th week (Graupe and Zhong ,2005), but this technique cant remove the noise or interference from the mixture. Many practical difficulties related with the ECG signal detection and extraction of fetus from mother ECG, which placed on the abdominal. Almost these due to low electrical signal which generated from fetus signal and contaminated by numerous signals of interference (Camps-Valls , 2004). Also the artifact signals represent one of the limitations in ECG signal analysis and can be solved by blind source extraction
1127
Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017 technique. The undesired signals should be removed from wanted ECG signal to get good diagnosis. Fuzzy and neural system is used to extract the fetus ECG from mother ECG signal (Saranya and Priyadharsini , 2010). Many algorithms and techniques for Fetus ECG detection and extraction presented with variable grades of achievement. Such as filters (Callaerts et.al., 1990), correlation (Camps-Valls et.al., 2004), Blind source extraction (BSE) (Jan. 2007), and many other techniques. Blind source separation or extraction is representing one of the most important and successful techniques used to extract the fetus ECG signals. However the ICA technique is not sufficient for acceptable fetus ECG signal extraction because the nonlinear association between the mother ECG and the fetus ECG signals due to mixing system inside the abdominal (Saranya and Priyadharsini, 2010; Abdullah et.al., 2014).
2. Theoretical background Different ways and algorithms to perform blind source extraction technique (BSE) such as dependent and independent component (DCA and ICA), Singular Value Decomposition (SVD), Common Spatial Pattern, and many another technique used to extract the useful information from mixture (Jan. 2007). EFICA and MCOMB techniques are used in this paper to get good separation process between the ECG signals without effect on the hidden information in the mixture. First of all, in any ICA technique if the sources have more Nongaussinity properties then the source represent more independency. In the model or formula the kurtosis mathematical operation is used to extent the Nongaussinity of the sources (Chaozhu et.al., 2013; Choung, 2013).
2.1 Kurtosis
It is a mathematical expression the 4th moment, generally, the 1st moment is the Mean value, the 2nd moment called the variance value, and the 3rd moment is identified as Skewness(Hippel, 2005). Kurtosis is used to investigate the random value that covers in a list of values. The most nongaussian value gives the independent sources as possible (Hyvarinen,1999).
2.2 Cross correlation (X-Correlation) X-correlation is used to identify the separated sources. Firstly, before any Xcorrelation process, the discrete Fourier transform (DFT) is used to transfer the source to time domain in order to make things easier in signal processing analyzing (Choung, 2013).
2.3 Independent Component Analysis (ICA) The ICA is used to analyze the sources into its components from their mixture. Generally the mixture can be classified into linear and nonlinear (Abdullah et.al., 2014). ICA is a multichannel process technique, where the detected signal is analyzed to list of ICs. The ICA technique is started for a cocktail party problem, this technique is suitable, and very useful for many applications in signal processing field. Figure 1 describes the main steps of ICA (Abdullah et.al., 2014; Abdullah et.al., 2014). S(t)
X(t)
A matrix
W matrix
Figure 1: Independent component Analysis Technique
1128
Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017
3. Fetal ECG signal The Fetus ECG electrical signals are associated to the mother ECG signal with similar waveform but weak amplitude and fast rate, it has P-wave, QRS-wave, and Twave as demonstrated in figure 2. These electric signals are generated by the reduction of the muscle of heart which named myocardium (Peddaneni ,2004; Hasan et.al., 2009).
Figure 2: Fetal ECG signal
4. Proposed work First of all, in order to explain the proposed algorithm, the conventional technique is illustrated in figure 3 to understand the behavior of previous techniques. For single fetal there is no more enhancement and difference between conventional and the proposed work. But in multi-fetal pregnancy, there are many advantages for blind source extraction technique compared with classical systems as demonstrated in the results(Graupe and Zhong ,2005).
Thoracic electrode
Adaptation algorithm _
H (Z)
∑
Estimated Signal for ECG
+ Abdominal electrode
Figure 3: Conventional method for M-ECG cancellation In this paper twins ECG signals (F1-ECG and F2-ECG) are measured from three sensors or electrodes placed on the chest and abdominal as shown in figure 4, first electrode called thoracic sensor and other two sensors called abdominal sensors (Fanaswala,2005).
Figure 4. Biological model of ECG signals 1129
Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017 In the conventional technique the thoracic sensor record only the mother ECG signal (M-ECG), but really it records mixture from three signals (M-ECG, F1-ECG, and F2-ECG), therefore this is one of the weak points in the conventional system. The abdominal sensors record mixtures from three sources (mother and fetus). The filtering technique can extract mother ECG signal from the fetus ECG signals, but cannot isolate F1-ECG from F2-ECG, therefore the blind source extraction technique is used here to solve this problem. Figure 5 shows the block diagram of the proposed algorithm, firstly, the signals are measured and acquisition and then initialize in order to prepare the signals for separation by EFICA technique to produce initial separation for the next step, the step 2 is MCOMB technique used to enhance the separation process and get optimum solution.
Measured signal
Data Acquisition
Step (1) EFICA
Initialization Centering
Whitening
Step (2) MCOMBI
Separated signals 𝑭𝟏 − 𝑬𝑪𝑮 𝑭𝟐 − 𝑬𝑪𝑮 𝑴 − 𝑬𝑪𝑮
Figure 5: Block diagram of the proposed algorithm
5. Simulation Results To check the performance of the proposed algorithm, MATLAB simulation is presented to simulate the ECG signals for mother (M-ECG) and twins (F1-ECG and F2ECG), the M-ECG signal is sampled by 4000 Hz with 72 beats/minute and amplitude about 3.5 mv. The fetus heart beats are faster than mother heartbeat as mentioned earlier in the introduction part. The rage of the rate about 139 beat/minute, weaker amplitude 0.25 mv. Figure 6 shows the simulated and generated ECG signals based on ECG( ) function.
(a)
(b)
(c)
Figure 6: input ECG signals: (a) mother ECG (b) Fetal-1 ECG (c) Fetal-2 ECG
Figure 7 demonstrates the mixture signals from three original sources mixed randomly, these mixtures represent the measured signals (raw data) from three electrodes
1130
Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017 placed on the chest and abdominal. The first step is to initialize the mixture of Preprocessing technique represented by centering and whitening.
Figure7: Mixture signals between mother ECG, Fetal-1 ECG, and Fetal-2 ECG
The first step after initialization process, the mixtures are separated by JADE technique but the results are not sufficient, therefore the MCOMB technique is used to enhance the separation process and get optimum separation as shown in figure 8.
Figure 8: Extracted ECG signals: (a) Extracted mother M-ECG (b) Extracted Fetal-1 ECG (c) Extracted Fetal-2 ECG
The simulation results are well, whatever the Interference signal ratio (ISR) measure is less as shown in Table 1. The results from this table is clearly better than JADE and MCOMBI techniques, the best average value of ISR for the proposed work is (-57.57 dB) and the minimum value is obtained by JADE about (-26.33 dB) as shown in the table 1.
1131
Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017 Table 1: Comparison of ISR values Methods
JADE MCOMBI Proposed
Interference signal ratio (ISR) for recovering signals S1 S2 S3 -22.13 -29.93 -50.90 -29.17 -56.28 -45.73 -46.91 -66.13 -59.67
ISR Mean -34.32 -43.72 -57.57
6. Conclusion In this work, the hybrid blind source extraction based on two BSE techniques is investigated to extract the fetus ECG signals (F1-ECG and F2-ECG) from mother ECG signal (M-ECG) and clean the signals from interference signals in order to not effect on the fetal health diagnosis. The performance of the proposed algorithm is checked and shown that the improvement enhancement in source separation. Generally, the proposed algorithm does not only suggest a novel algorithm, but furthermore offers a good judgment among the existing original BSE techniques in the works for twin ECG signal processing technique.
7. References Abdullah, A. K., Zhu Z. C., Abdullah A. A. A. and Siya L., 2014. "Automatic Extraction System for Common Artifacts in EEG Signals based on Evolutionary Stone’s BSS Algorithm." Mathematical Problems in Engineering: 25. Abdullah, A. K., Zhu Z. C., Siyao L. and Hussein S. M. ,2014. "Blind Source Separation Techniques Based Eye Blinks Rejection in EEG Signals." Information Technology Journal 13: 401-413. Álvarez, L. O. S. and Alberto Gonzalez J. M. 2015. Hybrid BSS Techniques for Fetal ECG Extraction Using a semi-synthetic database. Images and Computer Vision (STSIVA), 2015 20th Symposium on Signal Processing, , IEEE. Callaerts D., Moor B. D., Vandewalle J. and Sansen W. 1990. "Comparison of SVD methods to extract the fetal electrocardiogram from coetaneous electrode signals." Med. Biol. Eng. Comput. 28: 217 224. Camps-Valls G. 2004. " Foetal ECG Recovery using dynamic neural networks." Artificial Intelligence in Medicine 31(3): 197-209. Camps-Valls, G., Martínez-Sober M., Soria-Olivas E., Benedito R. M.-., Calpe-Maravilla J. and GuerreroMartínez J. ,2004. "Foetal ECG recovery using dynamic neural networks:." Artif. Intell. Med. 31(3): 197–209. Chaozhu, Z., Siyao L. and Abdullah A. K.,2013. A New Blind Source Separation Method to Remove Artifact in EEG Signals. Third International Conference on Instrumentation, Measurement, Computer, Communication and Control (IMCCC) IEEE. Choung, T. Y. ,2013. "Blind Signal Separation Implementation on Power Systems." EURECA 1(1). Fanaswala, M. 2005. "Fetal Electrocardiogram Extraction: A CaseStudy in Non-linear System Identification.".
1132
Journal of Babylon University/Engineering Sciences/ No.(4)/ Vol.(25): 2017 Gao, P., Chang E.-C. and Wyse L. ,2003. Blind separation of fetal ecg from single mixture using SVD and ICA. Proceedings of the 2003 Joint Conference of the Fourth International Conference on Information, Communications and Signal Processing., IEEE. Graupe, D., Zhong Y. ;M. H.,2005. "Extraction of Fetal ECG from Maternal ECG Early in Pregnancy." IJBEM 7(1). Hasan, M. A., Ibrahimy M. I. and Reaz M. B. I.,2009. "An Efficient Method for Fetal Electrocardiogram Extraction from the Abdominal Electrocardiogram Signal." Journal of Computer Science 5(9): 619-623. Hippel P. T. v. 2005. Mean, Median, and Skew: Correcting a Textbook Rule. Hyvarinen, A.,1999. "Fast and Robust Fixed-point algorithms for Independent Component Analysis." IEEE Transactions on Neural Networks Jan K. A. 2007. "Extraction of Fetal Electrocardiogram Using Adaptive Neuro-Fuzzy Inference Systems." IEEE Trans. Biomed. Eng. 54(1). Peddaneni h. 2004. Comparison of algorithms for fetal ecg extraction. Master of science, university of florida. Saranya.S and Priyadharsini S .S.; 2010. "A Novel Hybrid Soft Computing Technique for Extracting Fetal ECG from Maternal ECG Signal." International Journal of Computer Applications 3.
1133